arXiv:1908.05791v4 [hep-th] 4 May 2020 h lsia ii favre ler eeae yasubset a by generated algebra vertex a of limit classical the with lersi uhtplgclhlmrhcscosaddisc and sectors dimensional topological–holomorphic such sect in topological–holomorphic algebras possess dimensions, four in Abstract: iwnOh Jihwan field supersymmetric theories in algebras vertex Poisson JHEP to submission for Prepared 2 1 eateto hsc,Uiest fClfri,Berkeley California, of University Physics, of Department eiee nttt o hoeia hsc,Wtro,ON Waterloo, Physics, Theoretical for Institute Perimeter N uesmer ntredmnin n hoiswith theories and dimensions three in supersymmetry 2 = 1 , 2 ag ls fsprymti unu edtere,incl theories, field quantum supersymmetric of class large A N n uy Yagi Junya and uecnomlfil hoy h soitdPisnvert Poisson associated the theory, field superconformal 2 = 1 A970USA 94720 CA , 2 Y Canada 2Y5 N2L r.W omlt oso vertex Poisson formulate We ors. s oeeape.Frafour- a For examples. some uss flcloeaoso h theory. the of operators local of N supersymmetry 2 = dn l theories all uding xagbais algebra ex Contents

1 Introduction 1

2 Poisson vertex algebras in topological–holomorphic sectors 3 2.1 Topological–holomorphic sector 3 2.2 Topological–holomorphic descent 5 2.3 Secondary products 6 2.4 λ-bracket 7 2.5 Poisson vertex algebra 9 2.6 Dimensional reduction to the Poisson algebra of a TQFT 10 2.7 Lie algebra of zero modes 11

3 N = 2 supersymmetric field theories in three dimensions 11 3.1 Topological–holomorphic sector 11 3.2 Free chiral multiplets 12 3.3 Sigma models 16 3.4 Free vector multiplets 16 3.5 N = 2 superconformal field theories 17

4 N = 2 supersymmetric field theories in four dimensions 17 4.1 Topological–holomorphic sector 18 4.2 Vertex algebras for N = 2 superconformal field theories 20 4.3 Free hypermultiplets 24 4.4 Gauge theories 27

1 Introduction

In this work we study algebraic structures in topological–holomorphic sectors of super- symmetric quantum field theories, with one or more topological directions and a single holomorphic direction. These structures, known as Poisson vertex algebras [1–3], exist in N = 2 supersymmetric field theories in three dimensions and N = 2 supersymmetric field theories in four dimensions, among others. Our motivation for studying the Poisson vertex algebras comes from two main sources. One is the fact that local operators in topological quantum field theories (TQFTs) of cohomological type [4, 5] form Poisson algebras, some aspects of which were elucidated in the recent paper [6]. TQFTs being special cases of topological–holomorphic theories, it is natural to ask what the analogs of these Poisson algebras are in the topological–holomorphic setting. An answer is Poisson vertex algebras.

– 1 – Another source of motivation is the presence of vertex algebras in four-dimensional N = 2 superconformal field theories, which was uncovered in [7] and has been a subject of intensive research for the past several years. The classical limits of these vertex algebras are Poisson vertex algebras. The construction of Poisson vertex algebras we discuss in this paper indeed provides a bridge between the two lines of developments: whereas the Poisson vertex algebra for a four-dimensional unitary N = 2 superconformal field theory has a canonical deformation to the vertex operator algebra introduced in [7], dimensional reduction turns it into a Poisson algebra associated with a TQFT in one dimension less. We emphasize, however, that the construction itself requires no conformal invariance and applies to a broader class of theories. In section 2, we describe the structures that define a topological–holomorphic sector within a supersymmetric field theory, and demonstrate that local operators in this sector comprise a Poisson vertex algebra. Essential for the definition of a Poisson vertex algebra is the binary operation called the λ-bracket [8], which plays a role similar to the Poisson bracket in Poisson algebras. The λ-bracket is constructed via a topological–holomorphic analog of topological descent [4], much as the Poisson bracket between local operators in a TQFT is constructed via the descent procedure. This construction may be seen as providing a concrete physical realization of a special instance of the Poisson additivity theorem, proved by Nick Rozenblyum and also independently in [9].1 In section 3, we identify topological–holomorphic sectors in theories with N = 2 su- persymmetry in three dimensions, and determine the Poisson vertex algebras for chiral multiplets and vector multiplets. We content ourselves mostly with free theories here; to go beyond that one needs to incorporate quantum corrections, both perturbative and nonperturbative. In section 4, we consider Poisson vertex algebras for N = 2 supersymmetric field theo- ries in four dimensions. We show that if a theory is conformal, the associated Poisson vertex algebra is the classical limit of a vertex algebra, which is isomorphic to the vertex operator algebra of [7] in the unitary case. For free hypermultiplets, we confirm this relation by explicitly computing the Poisson vertex algebra and comparing it with the vertex operator algebra. For gauge theories constructed from vector multiplets and hypermultiplets, the relation to the vertex algebras allows us to propose a description of the associated Poisson vertex algebras. There are many directions to explore in connection with Poisson vertex algebras. To conclude this introduction we briefly describe some general ideas. Theories related by dualities, such as mirror symmetry and S-dualities, should have isomorphic Poisson vertex algebras. Dualities may thus lead to interesting isomorphisms between Poisson vertex algebras. Conversely, isomorphisms established between Poisson vertex algebras associated with apparently different theories may serve as evidence that those theories are actually equivalent. The structures considered in this paper can be further enriched by introduction of

1We are grateful to Dylan Butson for explaining this point to us.

– 2 – topological–holomorphic boundaries and defects, which themselves support Poisson vertex algebras (or vertex algebras if they are two-dimensional and have no topological direction). The interplay between the Poisson vertex algebras associated with the bulk theory and the boundaries and defects would be a fascinating topic. In three dimensions, relations between bulk Poisson vertex algebras and boundary vertex algebras are discussed in [10]. It would also be interesting to study the Poisson vertex algebras for theories originat- ing from compactification of six-dimensional N = (2, 0) superconformal field theories on Riemann surfaces [11, 12] and 3-manifolds [13–16]. These Poisson vertex algebras should be geometric invariants. Intriguing observations made in [17] suggest that the Poisson vertex algebras for four- dimensional N = 2 supersymmetric field theories are related to wall-crossing phenomena. Understanding this relation may be an outstanding problem. Lastly, it would be fruitful to investigate connections to integrable hierarchies of soliton equations. This is the context in which the theory of Poisson vertex algebras was originally developed [1–3, 18].

2 Poisson vertex algebras in topological–holomorphic sectors

In this paper we consider supersymmetric field theories that contain sectors that are topo- logical in d ≥ 1 directions and holomorphic in one complex direction. The topological i d directions form a d-dimensional manifold M, parametrized by real coordinates y = (y )i=1; the holomorphic direction is a Riemann surface C with complex coordinate z. We also use xµ, µ = i, z,z ¯, for the coordinates of M × C, with xi = yi, xz = z and xz¯ =z ¯. Typically, we take M = Rd and C = C or a cylinder. In this section we explain how such a topological–holomorphic sector arises from a supersymmetric field theory, and construct a (d-shifted) Poisson vertex algebra in this sector. The construction is similar to that of the Poisson algebras for TQFTs, treated in detail in [6]. Then, we discuss some of the basic properties of the Poisson vertex algebra thus obtained.

2.1 Topological–holomorphic sector Suppose that a quantum field theory on M × C is invariant under translations shifting xµ, and has symmetries generated by a fermionic conserved charge Q and a fermionic one-form conserved charge i Q = Qi dy + Qz¯ d¯z , (2.1) satisfying the relations

Q2 = 0 , (2.2)

[Q, Pµ] = 0 (2.3)

[Q, Pµ] = 0 , (2.4) i [Q, Q] = iPi dy + iPz¯ d¯z . (2.5)

– 3 – µ Here Pµ is the generator of translations in the x -direction, acting on a local operator O by2 iPµ · O = ∂µO . (2.6) The graded commutator [ , ] is defined by [a, b]= ab−(−1)F (a)F (b)ba, where (−1)F (a) = +1 if a is bosonic and −1 if fermionic. Since Q squares to zero, we can use it to define a cohomology in the space of operators. The commutation relations (2.3) and (2.5) show that Pµ act on the Q-cohomology, and Pi and Pz¯ act trivially. Therefore, the Q-cohomology class [[O(z)]] of a Q-closed local operator O(y,z, z¯) is independent of its position in M and varies holomorphically on C. The same is true for correlation functions of Q-closed local operators, which depend only on the Q-cohomology classes and not the choices of representatives. Given multiple Q-closed local operators Oa(xa), a = 1, . . . , n, located at xa = (ya, za, z¯a), we define the product of their Q-cohomology classes [[Oa(za)]] by

[[O1(z1)]] · · · [[On(zn)]] = [[O1(x1) ···On(xn)]] , xa 6= xb if a 6= b , (2.7) displacing the operators in M if necessary. (Note that this definition requires d ≥ 1.) The condition that no two operators are located at the same point in M × C is important because a product of coincident operators is generally singular in a quantum field theory. As long as no two operators are located at the same point in M, we may set all za equal in the above definition of product. Pointwise multiplication on C furnishes the Q-cohomology of local operators with the structure of an algebra, which is furthermore equipped with a derivation Pz. We denote this algebra by V. F While the space of operators is always Z2-graded by the fermion parity (−1) , in many examples this grading is refined by a Z-grading by a weight F of a U(1) symmetry such that F (Q)= −F (Q) = 1 , F (Pµ) = 0 . (2.8)

In that case, V is Z-graded by F . It may also happen that only a Z2n subgroup of such a U(1) symmetry is unbroken and V is Z2n-graded. There may be additional U(1) symmetries providing further gradings. We will refer to the collection of various structures associated with the Q-cohomology described above as the topological–holomorphic sector of the theory. This definition is somewhat weaker than the usual notion of a topological–holomorphic field theory of co- homological type, in which one requires that the components Tij and Tµz¯ = Tzµ¯ of the stress–energy tensor are Q-exact.

2In the path integral formulation, the action of a conserved charge X on a local operator located at O a point x is implemented by the integration of the Hodge dual of the associated conserved current over a simply-connected codimension-1 cycle surrounding x. If we take this cycle to be the difference of two time slices sandwiching x, we obtain the familiar formula X = [X, ] as an operator acting on the Hilbert ·O O space of states. The cycle can also be infinitesimally small, so the operation X is local. O 7→ ·O

– 4 – 2.2 Topological–holomorphic descent In cohomological TQFTs, there is a procedure called topological descent [4] which produces Q-cohomology classes of nonlocal operators starting from a Q-cohomology class of local operator. This construction can be adapted in the topological–holomorphic setting. For a local operator O, let us define its kth descendant O(k) by 1 O(k) = dxµ1 ∧···∧ dxµk (Q · · · Q ) · O . (2.9) k! µ1 µk ∗ Here we have set Qz = 0. We denote the total descendent of O by O :

d+1 O∗ = O(k) = exp(Q) · O . (2.10) Xk=0 Similarly, we introduce O−∗ = exp(−Q) · O . (2.11) These descendant operators satisfy

Q · O±∗ = ±d′O±∗ + (Q · O)∓∗ , (2.12) where O+∗ = O∗ and ′ i d =dy ∂i + d¯z∂z¯ . (2.13) Now suppose that O is a Q-closed local operator. Then, for any holomorphic one-form ω(z)= ωz(z)dz on C, we have

Q · (ω ∧ O∗)= −d(ω ∧ O∗) . (2.14)

Therefore, for any cycle Γ ⊂ M × C, the integral

ω ∧ O∗ (2.15) ZΓ is a Q-closed operator. The Q-cohomology class of this operator depends on Γ only through ∗ the homology class [[Γ]] ∈ H•(M × C) since d(ω ∧ O ) = 0 in the Q-cohomology. It also depends on O only through the Q-cohomology class [[O]] because shifting O by Q · O just −∗ changes the integral by Q · Γ ω ∧ O . We call this procedure of constructing nonlocal Q-cohomology classes from local ones topological–holomorphic descent. e R e More generally, for a product O1(x1) ···On(xn) of n Q-closed local operators, subject to the condition that xa 6= xb if a 6= b, we define its descendant as a differential form on the configuration space of n points on M × C,

n Confn(M × C)= {(x1,...,xn) ∈ (M × C) | xa 6= xb if a 6= b} . (2.16)

Let πa : Confn(M × C) → M × C be the projection to the ath factor, and for differential • forms α1, · · · , αn ∈ Ω (M × C) let

⊠ ⊠ ∗ ∗ • α1 · · · αn = π1α1 ∧···∧ πnαn ∈ Ω (Confn(M × C)) . (2.17)

– 5 – In this language, we may think of the product of local operators at n distinct points as the evaluation of a local operator at a single point in Confn(M × C):

O1(x1) ···On(xn) = (O1 ⊠ · · · ⊠ On)(x1,...,xn) . (2.18)

We define the descendant of O1 ⊠ · · · ⊠ On to be

⊠ ⊠ ∗ ∗ ⊠ F1 ∗ ⊠ ⊠ F1+···+Fn−1 ∗ (O1 · · · On) = O1 σ O2 · · · σ On , (2.19) where Fa = F (Oa) and the operator σ multiplies even forms by +1 and odd forms by −1. The descendant satisfies

∗ ′ ∗ Q · (O1 ⊠ · · · ⊠ On) =d (O1 ⊠ · · · ⊠ On) , (2.20)

′ • with d acting on Ω (Confn(M × C)) in the obvious manner. The role of σ in the defini- tion (2.19) is to supply, via the relation σd′ = −d′σ, a factor of (−1)F1+···+Fa−1 when d′ ∗ acts on Oa. n Given a holomorphic top-form ω on C (or more precisely, its pullback to Confn(M × C)) and a cycle Γ ⊂ Confn(M × C), the integral

∗ ω ∧ (O1 ⊠ · · · ⊠ On) (2.21) ZΓ is a Q-closed operator whose Q-cohomology class depends only on ω, the homology class [[Γ]] and the Q-cohomology classes [[Oa]].

2.3 Secondary products The algebra V of Q-cohomology of local operators possesses secondary products, in addi- tion to the ordinary (or primary) product given by pointwise multiplication on C. These products make use of topological–holomorphic descent.

Choose a translation invariant holomorphic two-form κ(z1, z2)= κz1z2 (z1, z2)dz1 ∧ dz2 on C × C. By translation invariance we mean

L∂z1 +∂z2 κ = 0 , (2.22) or (∂z1 + ∂z2 )κz1z2 (z1, z2) = 0, where LV denotes the Lie derivative with respect to V . For two Q-closed local operators O1 and O2, we define a new Q-closed operator O1 ⋆ O2 by

(d) (O1 ⋆κ O2)(x2)= ι∂z2 κ(z1, z2) ∧ O1 (x1) O2(x2) , (2.23) Sd+1 Z x2  d+1 where Sx2 is a (d + 1)-sphere centered at the point x2 at which O2 is placed, with the radius taken to be sufficiently small. On this operator iPµ acts as ∂µ since infinitesimally d+1 displacing O1 in the integrand is equivalent to shifting the center of Sx2 , which does not change the homology class, plus shifting the argument z2 in ι∂z2 κ(z1, z2). In particular, the relation (2.22) is crucial for the topological–holomorphic descent to hold for O = O1 ⋆κ O2.

– 6 – The Q-cohomology class [[O1 ⋆κ O2]] depends only on [[O1]] and [[O2]]. Hence, ⋆κ defines a secondary product for V. It has cohomological degree −d since O1 ⋆κ O2 uses the dth descendant of O1 which has fermion number F1 − d. The secondary product ⋆κ acts on the primary product as a derivation: it satisfies the Leibniz rule

(F1+d)F2 [[O1]] ⋆κ([[O2]][[O3]]) = ([[O1]] ⋆κ[[O2]])[[O3]]+ (−1) [[O2]]([[O1]] ⋆κ[[O3]]) . (2.24)

d+1 This identity simply says that S enclosing both O2 and O3 can be divided into two d+1 S , one containing O2 and the other containing O3. Such a decomposition is possible because O2 and O3 in the product [[O2]][[O3]] are by definition separated in M.

2.4 λ-bracket

d+1 Secondary products are defined locally; we can make Sx2 as small as we wish, for changing the radius does not affect its homology class. Then, by the locality of quantum field theory (expressed, for example, as the gluing axiom in the Atiyah–Segal formulation), any secondary product can be computed purely based on the knowledge of the behavior of the theory in a neighborhood U × V ⊂ M × C of the point at which we are taking the product, where U × V ≃ Rd × C topologically. On V , we can expand κ(z1, z2) around z2:

∞ 1 κ(z , z )= (z − z )k∂k κ (z , z )dz ∧ dz . (2.25) 1 2 k! 1 2 z1 z1z2 2 2 1 2 Xk=0 Since ⋆κ is linear in κ, all information about secondary products in the neighborhood U ×V k is encoded in the cases when κ(z1, z2) = (z1 − z2) dz1 ∧ dz2. The generating function of λ(z1−z2) such forms is e dz1 ∧ dz2, where λ is a formal variable. This consideration motivates us to introduce the λ-bracket [8]:

F1d λ(z1−z2) (d) {[[O1]] λ [[O2]]}(z2) = (−1) e dz1 ∧ O1 (x1) O2(x2) . (2.26) Sd+1 Z x2   The spacetime is here taken to have the topology of Rd × C, but not necessarily given the standard metric. In general, the λ-bracket may depend on the choice of metric and other geometric structures. The λ-bracket satisfies a few important identities, besides the Leibniz rule (2.24). First, it has sesquilinearity:

{∂z[[O1]] λ [[O2]]} = −λ{[[O1]] λ [[O2]]} , (2.27) {[[O1]] λ ∂z[[O2]]} = (λ + ∂z){[[O1]] λ [[O2]]} .

Second, it has the following symmetry under exchange of the two arguments:

(F1+d)(F2+d) {[[O1]] λ [[O2]]} = −(−1) {[[O2]] −λ−∂z [[O1]]} . (2.28)

– 7 – The λ-bracket with λ being an operator Λ is to be understood as a power series in Λ acting on the local operators in the bracket:

{[[O2]] Λ [[O1]]} = exp(Λ∂λ){[[O2]] λ [[O1]]} λ=0 . (2.29)
Finally, the λ-bracket satisfies the Jacobi identity

{[[O1]] λ {[[O2]] µ [[O3]]}}

(F1+d)(F2+d) = {{[[O1]] λ [[O2]]} λ+µ [[O3]]} + (−1) {[[O2]] µ {[[O1]] λ [[O3]]}} . (2.30)

Let us prove the above identities one by one. Since these are equalities between holo- morphic functions on C, it suffices to demonstrate that they hold on C× = C \ {0}. Thus we replace the spacetime with Rd × C× during the derivation. The sesquilinearity (2.27) is straightforward to show. The first equation follows from the fact that to the integral appearing in the definition (2.26) of the λ-bracket, an infinites- imal displacement of O1 in the z-direction has the same effect as shifting the parameter z2 d+1 λ(z1−z2) in Sx2 and e . To prove the second equation, one may just take a derivative of the λ-bracket with respect to z2. To establish the symmetry (2.28), we show

f(z2)dz2{[[O1]] λ [[O2]]}(z2) 1 ZS0 (F1+d)(F2+d) = −(−1) f(z1)dz1{[[O2]] −λ−∂z [[O1]]}(z1) (2.31) 1 ZS0 C× 1 for any holomorphic function f on , where S0 is a circle around z = 0. To this end we d × adopt the point of view of the configuration space. As an integral in Conf2(R × C ), the left-hand side is given by

F1d λz12 ∗ (−1) f(z2)dz2 ∧ e dz12 ∧ (O1 ⊠ O2) . (2.32) Sd+1×S1 Z x2 0  µ µ µ µ µ Rd C× Here we have defined x12 = x1 −x2 and think of (x12,x2 ) as coordinates on Conf2( × ). d+1 1 Rd C× d 1 d+1 3 The cycle Sx2 ×S0 ⊂ Conf2( × ) is homologous to (−1) S0 ×Sx1 , so we can rewrite this integral as

F1d+d −λz21 ∗ − (−1) f(z2)dz1 ∧ e dz21 ∧ (O1 ⊠ O2) . (2.34) S1×Sd+1 Z 0 x1  3 d × Since Hd+2(Conf2(R C )) = Z, to establish this relation we just need to evaluate a suitable cohomol- × ∼ −1 µ µ ogy class on these cycles. We can take this class to be d(θ1 + θ2) d (Vµ δ(x12)dx12), where θ = arg z and ∧ d × 2d+3 1 δ(x) is the delta function. To compute the homology group, let U = Conf2(R C ) and V R S × × × ≃ × be a normal neighborhood of the diagonal ∆ of (Rd C )2. Then, U V =(Rd C )2 R2d+2 S1 S1 × ∪ × ≃ × × and U V = V ∆ Rd+2 S1 Sd+1. From the Mayer–Vietoris sequence ∩ \ ≃ × ×

Hd+3(U V ) Hd+2(U V ) Hd+2(U) Hd+2(V ) Hd+2(U V ) , (2.33) ···→ ∪ → ∩ → ⊕ → ∪ →··· 1 d+1 we get Hd+2(U) = Hd+2(S S ) = Z. ∼ × ∼

– 8 – We expand f(z2) around z1 and replace the powers of z21 in the series with powers of −∂λ to get

∞ 1 F1d+d k k −λz21 ⊠ ∗ − (−1) (−∂λ) ∂z f(z1)dz1 ∧ e dz21 ∧ (O1 O2) k! S1×Sd+1 k=0 Z 0 x1  X ∞ F1d+d 1 k −λz21 ∗ = −(−1) (iPz∂λ) f(z1)dz1 ∧ e dz21 ∧ (O1 ⊠ O2) , (2.35) k! 1 d+1 S0 ×Sx1 Xk=0 Z  In the equality we have used the invariance of the Q-cohomology class of the integral under a shift of the cycle in the z-direction. The last expression is equal to

F1d+d+F1d+F1(F2+d)+F2d − (−1) f(z1)dz1 exp(iPz∂λ){[[O2]] −λ [[O1]]}(z1) , (2.36) 1 ZS0

F1d F1 ∗ where a factor of (−1) comes from σ in the definition (2.19) of (O1 ⊠ O2) , the factor F1(F2+d) (0) (d) F2d (−1) is due to an interchange of O1 and O2 , and a factor of (−1) comes from F2 ∗ σ in (O2 ⊠ O1) . This is what we wanted. The Jacobi identity (2.30), like the Leibniz rule, is based on the decomposition of a d+1 d+1 large S surrounding both O2 and O3 into a small S around O2 and another small d+1 d+1 S around O3. The latter small S gives the second term on the right-hand side. Obtaining the first term is a little tricky. The shift in µ is easy to understand. What ∗ ∗ ∗ ∗ is tricky is that this term involves (O1O2) , while the left-hand side only contains O1O2, ∗ which differs by terms involving Qµ acting on O1. To avoid getting these extra terms, we can use the symmetry (2.28). Let us switch to the configuration space viewpoint, as we did in the proof of the d × symmetry. The relevant cycle in Conf3(R × C ) is one that represents the situation in ∗ d+1 ∗ ∗ d+1 which O1 is integrated over S containing O2, while O2 is integrated over S around ∗ ∗ 1 C× O3, and finally O3 is integrated over S around the origin of . Up to a sign, this ∗ ∗ ∗ cycle is homologous to one in which O2 is surrounded by O1, and by O3 farther back, and itself circles around the origin of C×.4 Again up to a sign, the second cycle gives

{[[O3]] −λ−µ−∂z {[[O1]] λ [[O2]]}} and equals the term in question in the Jacobi identity. To determine the sign, we can go back to the original viewpoint and consider the special case λ(z1−z2) ∗ when Qµ annihilate d+1 e dz1 ∧ O . Sx2 1 2.5 Poisson vertexR algebra

The Q-cohomology of local operators V is a graded C[Pz]-module endowed with the λ- bracket { λ }: V⊗V→V[λ] with fermion number

F ({ λ })= −d , (2.37) satisfying the sesquilinearity (2.27), the symmetry (2.28) and the Jacobi identity (2.30). These properties make V a d-shifted Lie conformal algebra [8].

4 ∗ ∗ ∗ Think of 1 , 2 and 3 as the moon, the earth and the sun, respectively, and consider the heliocentric O O O × versus geocentric descriptions of their relative motion. The origin of C can be the center of our galaxy.

– 9 – In addition, the λ-bracket satisfies the Leibniz rule (2.24). A unital commutative associative d-shifted Lie conformal algebra, with the Leibniz rule obeyed, is called a d-shifted Poisson vertex algebra. Thus, we have shown that the algebra of local operators in the topological–holomorphic sector of a quantum field theory on M ×C possesses the structure of a d-shifted Poisson vertex algebra. We refer the reader to [19] for an introduction to Poisson vertex algebras. Suppose that the theory has rotation symmetry on C, and let J be its generator. We normalize J in such a way that Pz has J = 1. If Q has a definite spin J(Q) (as opposed to being a linear combination of operators with different spins), then V, as an algebra, is also graded by J. In this case we can choose Q to have spin J(Q)= −J(Q) . (2.38) As it is, the λ-bracket is not compatible with this grading. The problem is that for computation of the λ-bracket, one of the local operators needs to be displaced from the d+1 center of rotations x2, and while [[Sx2 ]] used in the λ-bracket is invariant under rotations, the factor eλ(z1−z2) is not. This means that the λ-bracket, when expanded in λ, consists of infinitely many parts carrying different values of J. We can, however, remedy this problem if we simultaneously shift the phase of λ. In other words, the λ-bracket respects the J-grading if we assign J(λ) = 1 . (2.39)

Since its definition contains the one-form dz1 and d copies of Q, the λ-bracket changes the J-grading by J({ λ })= −J(Q)d − 1 . (2.40)

2.6 Dimensional reduction to the Poisson algebra of a TQFT Poisson vertex algebras are generalizations of Poisson algebras, which are commutative associative algebras endowed with a Lie bracket obeying the Leibniz rule. Indeed, the quotient of V by the ideal (∂zV)V is a d-shifted Poisson algebra with respect to the Poisson 0 bracket induced from the λ -order part { λ }|λ=0 of the λ-bracket. This Poisson algebra may be interpreted as a structure associated with the (d + 1)- dimensional theory that one obtains by taking C to be a cylinder and performing dimen- sional reduction on the circumferential direction. Since local Q-cohomology classes vary holomorphically on C, after the reduction they become independent of their positions in the longitudinal direction. Thus, the topological–holomorphic sector on M × C is turned into a topological sector on M × R. The latter carries a Poisson bracket of degree −d on the algebra of local operators [6], and in favorable situations, V/(∂zV)V coincides with this Poisson algebra. In general, the relation between the two Poisson algebras can be more complicated. For instance, the theory on M × R may have local operators that come from line operators in M × C extending in the reduced direction. If these local operators are Q-closed, they may represent Q-cohomology classes that are not present in V/(∂zV)V. If they are not Q- closed, then some elements of V/(∂zV)V may be paired up with them and get annihilated from the Q-cohomology.

– 10 – 2.7 Lie algebra of zero modes

As a vector space, we can also consider the quotient V = V/∂zV. This is the space of zero modes of local Q-cohomology classes. The quotient map V → V is denoted by . The space V is actually a d-shifted Lie algebra, with the Lie bracket { , } of degree R −d given by

{ [[O1]], [[O2]]} = {[[O1]] λ [[O2]]}|λ=0 . (2.41) Moreover, the Lie algebra VR acts onR V by R

{ [[O1]], [[O2]]} = {[[O1]] λ [[O2]]}|λ=0 . (2.42)

As such, V may be regardedR as a Lie algebra generating a continuous symmetry of V. A proof of these statements is straightforward [19].

3 N =2 supersymmetric field theories in three dimensions

The lowest dimensionality in which the structures of topological–holomorphic sectors arise is three. In this section, we define Poisson vertex algebras for three-dimensional N = 2 supersymmetric field theories and determine them in basic examples. We will take

M × C = R × C (3.1) and denote the coordinate on R by t.

3.1 Topological–holomorphic sector

The N = 2 supersymmetry algebra in 2 + 1 dimensions has four supercharges Qα, Qα, α = ±, satisfying the commutation relations5

[Q±, Q±]= −P0 ± P2 , (3.2)

[Q±, Q∓]= P1 ∓ iZ , (3.3)

[Qα,Qβ] = [Qα, Qβ] = 0 , (3.4) where Z is a real central charge. Performing the Wick rotation x0 →−ix3 and introducing the complex coordinate z = (x2 + ix3)/2, the commutation relations (3.2) become

[Q+, Q+]= Pz , [Q−, Q−]= −Pz¯ . (3.5)

Taking the x1-direction as the topological direction M = R and the z-plane as the 2 holomorphic direction C = C, we wish to find a supercharge Q such that Q = 0 and Pt, Pz¯ are Q-exact. Moreover, Pz should not appear in any commutators between Q and other

5The = 2 supersymmetry algebra in 2+1 dimensions can be obtained from the = 1 supersymmetry N N algebra in 3+1 dimensions by dimensional reduction. For the latter algebra we follow the conventions of [20], except that we rescale the supercharges by a factor of 1/√2. We have chosen to perform the reduction along the x2-direction and subsequently renamed x3 to x2.

– 11 – generators so that it is unconstrained in the Q-cohomology. These conditions require Q to take the form

Q = aQ− + dQ− , (3.6) 2 with a, d being complex numbers. We have Q = −adPz¯, so we must set either a or d to zero. Without loss of generality, we can take

Q = Q− . (3.7)

From the commutation relations (3.2)–(3.4), we see that in general Pt cannot be gen- uinely Q-exact; rather, the combination Pt − iZ is Q-exact. An obvious way to construct a Poisson vertex algebra in such a situation is to take the Q-cohomology in the subalgebra of local operators that have Z = 0. Instead, we may simply regard Pt − iZ as generating a modified translation symmetry in the t-direction, which is “twisted” by the central charge. Given a local operator O, we define its twisted-translated counterpart Z O by

Z O(t,z, z¯) = exp i(Pt − iZ)t · O(0, z, z¯) = exp(Zt) · O(t,z, z¯) . (3.8)

Z Then, the Q-exact operator i(Pt − iZ) acts
on O as ∂t. We can thus construct a Poisson vertex algebra from local operators translated by Pt − iZ. In many cases, there is a U(1) R-symmetry under which Qα has charge R = −1 and Qα has R = 1. Requiring R(Q)= −R(Q) = 1 fixes the one-form supercharge:

Q = iQ+dt − iQ−d¯z . (3.9)

We have J(Q)= −J(Q)= −1/2, so the λ-bracket has (R, J) = (−1, −1/2). It is often useful to twist the theory by regarding R J ′ = J + (3.10) 2 as a generator of rotations on C. Under the twisted rotations, J ′(Q) = J ′(Q) = 0, which means that Q is a scalar and Q transforms correctly as a differential form. We have

′ J ({ λ })= −1 . (3.11)

3.2 Free chiral multiplets Let us determine the Poisson vertex algebra for a free theory of chiral multiplets. This is arguably the simplest N = 2 supersymmetric field theory. A chiral multiplet Φ consists of a complex bosonic scalar φ, a pair of fermionic spinors

ψ± and a complex bosonic scalar F . Under the action of the linear combination ǫ−Q+ − ǫ+Q− − ǫ¯−Q+ +ǫ ¯+Q− of the supercharges, these fields transform as

δφ = ǫ−ψ+ − ǫ+ψ− ,

δψ+ = i¯ǫ−∂zφ − i¯ǫ+(∂t + m)φ + ǫ+F , (3.12) δψ− = i¯ǫ−(∂t − m)φ + i¯ǫ+∂z¯φ + ǫ−F ,

δF = −i¯ǫ+ ∂z¯ψ+ + (∂t + m)ψ− − i¯ǫ− (∂t − m)ψ+ − ∂zψ− ,

– 12 – where m is a real mass parameter called the twisted mass. The supersymmetry transfor- mations for the conjugate antichiral multiplet Φ = (φ,¯ ψ¯±, F ) can be obtained by hermitian conjugation, which exchanges unbarred fields to the corresponding barred fields and vice † † versa. (Note that ∂z = ∂z and ∂z¯ = ∂z¯ because the hermitian conjugation is defined with respect to Minkowski spacetime.) The central charges of Φ and Φ are given by their masses:

Z(Φ) = m , Z(Φ) = −m . (3.13)

a a a a We consider a theory described by n chiral multiplets Φ = (φ , ψ±, F ), a = 1, . . . , n, a¯ ¯a¯ ¯a¯ a¯ and their conjugates Φ = (φ , ψ±, F ), with no superpotential. The action of the theory is

3 a ¯¯b d x(Q−Q−Q+Q+) · (ga¯bφ φ ) R C Z × ¯ 3 a ¯¯b a ¯¯b a b = d x ga¯b ∂z¯φ ∂zφ + (∂t + ma)φ (∂t + mb)φ − F F R×C Z  a a ¯¯b a ¯¯b a ¯¯b + i ∂z¯ψ+ + (∂t + ma)ψ− ψ+ + iψ−∂zψ− − iψ+(∂t + mb)ψ− , (3.14) 
a a a a where ga¯b = δab. We assign (R, J) = (0, 0) to Φ . Then, φ , ψ± and F have (R, J) = (0, 0), (−1, ±1/2) and (−2, 0), respectively, and (−1)F = (−1)R. To deal with the Q-cohomology of this theory, it is better to switch to a notation suitable for twisted theories. Under the twisted rotations on C, all fields transform like components of differential forms, which we can make manifest by renaming them as follows:

a a a a a χ = iψ+dt − iψ−d¯z , G = F dt ∧ d¯z , ¯ (3.15) a¯ ¯a¯ ¯ ¯¯b b η¯ = ψ− , ξa = ga¯bψ+dz . Ga = ga¯bF dz .

The supercharge Q = Q− acts on the fields as

Q · φa = 0 , Q · φ¯a¯ =η ¯a¯ , a ′ a a¯ Q · χ =dm φ , Q · η¯ = 0 , a (3.16) a ′ a ¯ Q · G =dma χ , Q · ξa = Ga ,

Q · Ga = 0 .

Here ′ dm =dt(∂t + m)+d¯z∂z¯ (3.17) is the d′-operator twisted by mass m. The action (3.14) can be written as

a ¯¯b a ¯ Q · ga¯bχ ∧ ⋆ dt(∂t + mb)+dz∂z φ + 2iG ∧ ξa . (3.18) R C Z × 
 The Poisson vertex algebra V is the Q-cohomology taken in the space of twisted- translated local operators. In terms of these operators the twisted masses disappear from the Q-action: Q · Z χa =d′Z φa , Q · ZGa =d′Z χa . (3.19)

– 13 – It follows that the Q-cohomology is independent of the twisted masses. (The dependence of the action on the twisted masses is Q-exact.) This is a consequence of the locality of the Poisson vertex algebra: the twisted masses may be thought of as vacuum expectation values of scalars in nondynamical vector multiplets for flavor symmetries and, as such, are determined by the boundary condition at infinity, on which V does not depend. We will set ma = 0 below to simplify our argument. Not all local operators are relevant for the computation of V. Suppose that we rewrite the action in such a way that it couples to the product metric on R × C in a manner that is invariant under diffeomorphisms on R and those on C. (This is achieved by introduction of ∗ a ¯¯b the topological term R×C dt∧φ ωCn to the action, where ωCn = iga¯bdφ ∧dφ is the K¨ahler form on Cn.) If V is a vector field generating such diffeomorphisms, J = V µT dxν is R V µν a current such that an integral of ⋆ JV acts on local operators by V . The components Ttt and Tz¯z¯ of the stress–energy tensor are Q-exact because the action is Q-exact and variations of the corresponding components of the metric commute with the Q-action, while Ttz = Ttz¯ = 0 as the metric is of a product form and Tzz¯ = 0 by conformal invariance on C. As a result, the diffeomorphisms on R and antiholomorphic reparametrizations on C act on the Q-cohomology trivially. In particular, Q-closed operators that have nonzero scaling dimension on R or nonzero antiholomorphic scaling dimension on C are Q-exact. a a¯ a¯ The relevant operators are therefore local operators constructed from φ , φ¯ ,η ¯ , ξ¯a and Ga, as well as their z-derivatives. Among these, Ga are auxiliary fields and vanish by ¯b ¯ k a¯ equations of motion. Since ga¯b∂zη¯ = ∂tξaz by an equation of motion, ∂z η¯ with k ≥ 1 k a¯ k ¯ k ¯a¯ can be dropped. In the absence of ∂z η¯ and ∂t ξaz, operators that contain ∂z φ cannot be k ¯a¯ Q-closed. Thus we can also drop ∂z φ . We may think of (φa, φ¯a¯) as coordinates on the target space

X = Cn (3.20)

a¯ ¯a¯ k a and identifyη ¯ with dφ . Furthermore, for each k ≥ 1, we may regard ∂z φ as a section of ∨ k−1 the holomorphic cotangent bundle T , and ∂ ξ¯az as a section of the tangent bundle TX . X z ′ ′ ′ 0,• j Under this identification, the relevant operators with J = j are sections of ΩX ⊗ EX , j′ where EX is a holomorphic vector bundle given by the formal series

∞ ∞ ∞ ∞ J′ J′ lk l ∨ mk m u EX = u S TX ⊗ u Λ TX . (3.21) ′ m ! JM=0 Ok=1 Ml=0  M=0  l ∨ For each k, the factor of the lth symmetric tensor power S TX accounts for operators that ¯ k a1 k al m take the form f(φ, φ)a1...al ∂z φ · · · ∂z φ , and the mth exterior power Λ TX accounts for ¯ a1...am k−1 ¯ k−1 ¯ those of the form f(φ, φ) ∂z ξa1z · · · ∂z ξamz. Using the equations of motion Ga = 0, we see that Q acts on these sections as the Dolbeault operator ∂¯, increasing R by 1 but keeping J ′ unchanged. We conclude that the Q-cohomology of local operator is isomorphic to the Dolbeault cohomology of this bundle:

∞ J′ J′ ∼ • J′ V = V , V = H∂¯(X; EX ) . (3.22) ′ JM=0

– 14 – Cn q J′ ¯ For X = , we have H∂¯(X; EX )=0 for q ≥ 1 by the ∂-Poincar´elemma. Therefore, V is • isomorphic as an algebra to the algebra of holomorphic sections of EX . The λ-bracket has (R, J ′) = (−1, −1). There are three combinations of local oper- a b ators for which we need to compute the λ-bracket, namely {[[φ ]] λ [[φ ]]}, {[[ξ¯az]] λ [[ξ¯bz]]} b and {[[ξ¯az]] λ [[φ ]]}; the other combinations can be obtained from these by the Leibniz rule, sesquilinearity and symmetry of the λ-bracket. Since there are no Q-cohomology classes with R< 0, we immediately find

a b {[[φ ]] λ [[φ ]]} = 0 . (3.23)

In general, {[[O1]] λ [[O2]]} vanishes unless the dth descendant of O1 produces a singularity d+1 when placed at the same point as O2, for otherwise the integration cycle Sx2 can be shrunk to a point. The theory under consideration is a free theory, so the product of a a b bosonic field and a fermionic one, such as those that appear in {[[φ ]] λ [[φ ]]}, cannot be singular. For the same reason we have

{[[ξ¯az]] λ [[ξ¯bz]]} = 0 . (3.24)

b ′ The remaining combination {[[ξ¯az]] λ [[φ ]]} is a polynomial in λ with (R, J ) = (0, 0). Since J ′(λ) = 1 and there are no local Q-cohomology classes with J ′ < 0, it must be actually independent of λ and equal to a Q-cohomology class with (R, J ′) = (0, 0), that is, a holomorphic function of φa. The theory has a U(n) global symmetry under which φa transform in the vector representation and ξ¯a in the dual representation, and the λ-bracket must be invariant under this symmetry. These constraints leave the only possibility to be ¯ b b that {[[ξaz]] λ [[φ ]]} is proportional to δa. Let us calculate this λ-bracket explicitly. For this calculation we restore the twisted masses to demonstrate that they do not affect the result. The first descendant of ξ¯az is given by

¯ ¯¯b Q · ξaz = ga¯b dt∂z − d¯z(∂t + ma) φ , (3.25) so we have

b λz −mat c¯ b {[[ξ¯az]] λ [[φ ]]}(0) = − e dz ∧ e gac¯ dt∂z − d¯z(∂t + ma) φ¯ (x) φ (0) (3.26) 2 ZS0  
where the factor of e−mat comes from twisted translation. Using the Stokes theorem we 3 2 can convert this integral to one over the 3-ball B0 that bounds S0 : i 3 λz−mat 2 2 ¯c¯ b − d x e gac¯(∂t + ∂z∂z¯ − ma)φ (x) φ (0) . (3.27) 2 3 ZB0   By integration by part in the path integral we obtain the identity

δφb(0) δS = φb(0) , (3.28) δφa(x) δφa(x)

– 15 – which holds inside correlation functions as long as no other operators are present at x. The b left-hand side is δa times the delta function supported at x = 0, whereas the right-hand 2 2 ¯c¯ b side is −gac¯(∂t + ∂z∂z¯ − ma)φ (x)φ (0). Plugging this relation into the integral (3.27), we find i {[[ξ¯ ]] [[φb]]}(0) = δb . (3.29) az λ 2 a b As expected, the result is proportional to δa, with the proportionality constant independent of the twisted masses. Finally, let us consider the Poisson algebra V/(∂zV)V associated with the dimensional reduction of the theory. The elements of V/(∂zV)V are in one-to-one correspondence with a the local operators constructed from φ and ξ¯az but not their z-derivatives, that is, the holo- • morphic sections of Λ TX . The Poisson bracket on V/(∂zV)V coincides with the Schouten– Nijenhuis bracket. This Poisson algebra is the one for the B-model with target space X [6].

3.3 Sigma models The construction of the Poisson vertex algebra discussed above can be generalized to the case when the target space X is not Cn but any other K¨ahler manifold. Thus we get a map X 7→ V(X) (3.30) which assigns to a K¨ahler manifold X a Poisson vertex algebra V(X). Classically, V(X) is still described by the Dolbeault cohomology of X with values in the holomorphic vector bundles (3.21). Quantum corrections may alter this description, however.

3.4 Free vector multiplets A theory of free vector multiples provides an example of a sigma model with X 6= Cn. In three dimensions, an abelian gauge field A can be dualized to a periodic scalar γ through the relation dγ = i ⋆ dA (in Euclidean signature). Under this dualization process a vector multiplet is mapped to a chiral multiplet whose scalar field is φ = σ + iγ, where σ is the adjoint scalar in the vector multiplet. Hence, a theory of n abelian vector multiplets is equivalent to a sigma model with target X = (R × S1)n. The target space metric enters the action through Q-exact terms, so we can actually take X = (C×)n , (3.31) with the standard flat metric on it. The target being flat, the theory is free and there are no quantum corrections to the description of the Poisson vertex algebra as the Dolbeault cohomology. Compared to the case of X = Cn, we have more ingredients to build Q-cohomology • a a¯ ¯a¯ classes from: holomorphic sections of the bundles EX can have poles at φ = 0, andη ¯ /φ are Q-closed but not Q-exact. Accordingly, we have more combinations to consider for the λ-bracket. Specifically, we have the cases when either operator in the λ-bracket involves [[¯ηa¯/φ¯a¯]], a ¯b ¯b a¯ a¯ ¯b ¯b ¯b ¯b and need to evaluate {[[φ ]] λ [[¯η /φ¯ ]]}, {[[¯η /φ¯ ]] λ [[¯η /φ¯ ]]} and {[[ξ¯az]] λ [[¯η /φ¯ ]]}. However,

– 16 – these additional combinations all vanish. For the first two, this is because there are simply no Q-cohomology classes with J ′ = −1. The last one has (R, J ′) = (1, 0) and may be proportional to [[¯ηa¯/φ¯a¯]]. This vanishes since neitherη ¯a¯ nor φ¯a¯ produces a singularity when (1) multiplied by ξ¯az .

3.5 N = 2 superconformal field theories Although the Poisson vertex algebra of an interacting N = 2 supersymmetric field theory is difficult to determine, some general statements can be made about it if the theory is unitary and has conformal symmetry. In that case, by conformal symmetry the Q-cohomology of local operators is isomorphic as a vector space to the Q-cohomology of states in radial quantization, and by unitarity the latter is isomorphic to the space of Q-harmonic states on which {Q,Q†} = 0. In radial quantization, the hermitian conjugate of Q is a conformal supercharge and satisfies

[Q,Q†]= D − R − J , (3.32) where D is the dilatation operator. Hence, a local Q-cohomology class is represented uniquely by a local operator with D − R − J = 0. If we assign D(λ) = 1, the λ-bracket has (D,R,J) = (−3/2, −1, −1/2) and preserves D − R − J. There are many N = 2 superconformal multiplets that contain such local operators; see [21] for a comprehensive list. Especially important are those with conserved currents. For example, there is a flavor current multiplet which contains a fermionic operator (1) Ψ0 with (D,R,J) = (3/2, 1, 1/2). The first descendant Ψ0 of this operator is given by (1) dz ∧ Ψ0 = ⋆(j − jz¯d¯z), where j is the conserved current for a flavor symmetry. The component jz¯ is Q-exact, so the zero mode [[Ψ0]] acts on V by an infinitesimal flavor transformation. For the theory of free chiral multiplets, (Ψ )a = ξ¯ φa. R 0 b bz Similarly, for each integer n ≥ 1, there is a multiplet that contains conserved currents and a local operator Ψn with (D,R,J) = (n/2 + 3/2, 1, n/2 + 1/2) representing a Q- (1) cohomology class. For n = 1, the first descendant Ψ1 is part of a spin-3/2 current, and [[Ψ1]] acts by a supersymmetry transformation. The multiplet with n = 2 contains a stress–energy tensor, and [[Ψ ]] acts by the holomorphic derivative ∂ . For free chiral R 2 z multiplets, (Ψ )a = ξ¯ ∂ φa. The multiplets with n ≥ 3 contain higher spin currents. 2 b bz z R

4 N =2 supersymmetric field theories in four dimensions

Next, we turn to Poisson vertex algebras for four-dimensional supersymmetric field theo- ries. In four dimensions, we must have twice as many topological directions as in three dimensions. It turns out that we also need twice as many supercharges, namely eight supercharges, generating N = 2 supersymmetry. We will take

M × C = R2 × C . (4.1)

– 17 – 4.1 Topological–holomorphic sector

A The supercharges Qα , QAα˙ , A = 1, 2, α = ±, of the N = 2 supersymmetry algebra in 3 + 1 dimensions satisfy the commutation relations

A µ A [Qα , QBα˙ ]= σαα˙ PµδB , (4.2) A B AB [Qα ,Qβ ]= ǫαβǫ Z , (4.3)

[QAα˙ , QBβ˙ ]= −ǫα˙ β˙ ǫABZ , (4.4) where Z = Z† is a complex central charge. Our convention is such that

−1 0 0 1 0 −i 1 0 σ0 = , σ1 = , σ2 = , σ3 = (4.5) 0 −1 1 0 i 0 0 −1        

12 A AB and ǫ = −ǫ12 = −ǫ+− = −ǫ+˙ −˙ = 1. We raise and lower indices as Qα˙ = ǫ QBα˙ , B QAα˙ = ǫABQα˙ . We perform the Wick rotation x0 → −ix4 and introduce the complex coordinates 1 2 3 4 µ w = (x + ix )/2 and z = (x + ix )/2. The generators Pαα˙ = σαα˙ Pµ of translations are then given by P P P P ++˙ +−˙ = z w . (4.6) P P P −P  −+˙ −−˙   w¯ z¯  We choose the w-plane for M = R2 and the z-plane for C = C. 2 We need a supercharge Q such that Q = 0 and P+−˙ , P−+˙ , P−−˙ are Q-exact. Since Pz should not enter any Q-commutators, Q must be of the form

1 2 Q = aQ− + bQ− − cQ2−˙ + dQ1−˙ . (4.7)

We have 2 a b Q = − det(A)Pz , A = . (4.8) ¯ c d   For Q2 = 0, the two rows of the matrix A must be proportional to each other. On the other hand, the Q-commutators of supercharges that contain P+−˙ and P−+˙ can be written as Q2 −Q1 Z −P [Q, + + ]= w A . (4.9) Q Q P Z  1+˙ 2+˙   w¯  (The commutator on the left-hand side is to be understood as acting on each entry of the matrix.) For Pw and Pw¯ to both appear on the right-hand side, neither row of A should be zero. Thus we find 1 2 Q = a(Q− + tQ2−˙ )+ b(Q− − tQ1−˙ ) (4.10) for some t 6= 0. From the commutators (4.9) we see that the linear combinations

−1 Pw + t Z , Pw¯ − tZ (4.11)

– 18 – are Q-exact. As in the case of three-dimensional N = 2 supersymmetry, we define twisted- translated local operators by

t,Z −1 O(w, w,z,¯ z¯) = exp i(Pw + t Z)w + i(Pw¯ − tZ)w ¯ · O(0, 0, z, z¯) −1 (4.12) = exp(i t Zw − itZw¯) · O(w, w,z,¯ z¯)

−1 t,Z so that i(Pw + t Z) and i(Pw¯ − tZ) act on O as ∂w and ∂w¯, respectively. Twisted- translated local Q-cohomology classes form a Poisson vertex algebra V.6 1 2 Let us suppose that the theory has an R-symmetry SU(2)R under which (Qα,Qα) 1 2 and (Qα˙ , Qα˙ ) transform as doublets. Then, the structure of V does not depend on the parameters a, b even though they appear in the formula (4.10) for Q. Under the action of the complexification SL(2, C)R of the R-symmetry group, V gets mapped to an iso- morphic algebra. Since the linear combinations of supercharges in the parentheses in the expression (4.10) form a doublet of SU(2)R, by the transformation by

0 −b ∈ SL(2, C)R (4.13) b−1 a   we can set 1 Q = Q− + tQ2−˙ (4.14) (assuming b 6= 0 without loss of generality). With this choice, Q has charge R = 1 under the diagonal subgroup U(1)R of SU(2)R. Under the rotation symmetry on C, it has spin J(Q)= −1/2. R2 1 Also, V is independent of t if the theory has rotation symmetry on . Since Q− and Q2−˙ have spin J = −1/2 and 1/2 under the rotation symmetry, we can change the value of t using complexified rotations; in effect, t transforms as if it has weight 1 under this × C -action, juste as ∂w does. Since rotations act on the space of local operators at w = 0, it follows that the Q-cohomology of local operators, as a vector space, is independent of t at w = 0, hence anywhere on R2. Moreover, by the action of the rotation symmetry, we can also show that the algebra structure and the λ-bracket remain the same under phase rotation of t.7 These structures depend holomorphically on t, so they are actually entirely independent of t. Thus, we can take t = 1 (4.15) and 1 Q = Q− + Q2−˙ . (4.16) We will write Z O for 1,Z O:

Z O(w, w,z,¯ z¯) = exp(iZw − iZw¯) · O(w, w,z,¯ z¯) (4.17)

6The fact that the Q-cohomology of local operators is a Poisson vertex algebra was mentioned in [22–24]. 7Here we cannot use the C×-action because we need to place local operators away from w = 0 in order to define these structures. The action by α C× transforms t,Z (w, w,z,¯ z¯) to αt,Z (αw, α−1w,z,¯ z¯) ∈ O− O (assuming, for simplicity, that is a scalar operator). For αw and α 1w¯ to be complex conjugate to each O other, we must have α U(1) or w = 0. ∈

– 19 – The one-form supercharge is not uniquely determined. The general form of Q that has (R, J) = (−1, 1/2) is

2 2 Q = iQ+dw + iQ1+˙ dw ¯ + i (u − 1)Q− − uQ1−˙ d¯z , (4.18) with u being an arbitrary complex number. The λ-bracket does not
depend on u, however. To see this, pick a boundary condition such that all local Q-cohomology classes vanish at R2 3 the infinity of . Furthermore, we choose to represent the homology class [[Sx2 ]] in the R2 1 R2 C definition (2.26) of the λ-bracket by a “cylinder” whose “side” is × Sz2 ⊂ × , where 1 R2 Sz2 has a fixed radius everywhere except at the infinity of where it shrinks to a point; 3 by the Poincar´econjecture, this cylinder is homeomorphic to Sx2 . Then, Qz¯ drops out of the computation of the λ-bracket because the only contributions come from the region in which the pullback of dz ∧ d¯z vanishes.

4.2 Vertex algebras for N = 2 superconformal field theories If the theory has not only N = 2 supersymmetry but also conformal symmetry, V can be deformed to a family of vertex algebras V ~ parametrized by ~ ∈ C. This is essentially quantization of V by Ω-deformation [25, 26], and related via dimensional reduction to the quantization of a Poisson algebra associated with an N = 4 superconformal field theory in three dimensions [6, 27]. α A An N = 2 superconformal field theory has eight conformal supercharges SA, Sα˙ , in A addition to the Poincar´esupercharges Qα , QAα˙ . In radial quantization the two sets of supercharges are related by hermitian conjugation:

A † α † Aα˙ (Qα ) = SA , (QAα˙ ) = S . (4.19)

A A Furthermore, the theory has an extra R-symmetry U(1)r, under which Qα , Sα˙ have charge α r = 1/2 and QAα˙ , SA have r = −1/2. The central charge Z necessarily vanishes because of U(1)r. Let us introduce the following deformations of P , Q and Q:

~ 2 P = Pwdw + Pw¯dw ¯ + Pzdz + (Pz¯ − ~R 1)d¯z , (4.20) ~ 1 ~ 2−˙ − Q = Q− + Q2−˙ + (S − S1 ) , (4.21) ~ Q = Q , (4.22)

A 1 2 Here R B are the generators of SU(2)R × U(1)r =∼ U(2), in terms of which R = R 1 − R 2 1 2 and r = R 1 + R 2. These deformed generators satisfy the relations

~ (Q )2 = ~(J + r) , (4.23) ~ ~ ~ ν ~ [Q , Pµ ]= −i ∂µJ Qν , (4.24) ~ ~ e [Q , Pµ ] = 0 , (4.25) ~ ~ ~ ie ~ [Q , Q ] = iPi dy + iPz¯ d¯z . (4.26)

– 20 – Recall that J is the generator of rotations on R2; it is normalized in such a way that [J, Pw]= Pw and [J, Pw¯ ]= −Pw¯. We have also denoted the corresponding vector field by the same symbol:e e e J = w∂w − w∂¯ w¯ . (4.27) ~ We regard P as a twisted translatione generator and define twisted-translated local oper- ators by

~ ~ ~ O (w, w,z,¯ z¯) = exp(izPz + i¯zPz¯ ) · O(w, w,¯ 0, 0) (4.28)

~ ~ so that iPµ act on O as ∂µ. The deformed supercharge Q~ does not square to zero, but instead to the generator

J ′ = J + r (4.29)

~ of twisted rotations on R2. (In this senseeQ definese an Ω-deformation [28, 29] and induces quantization [30].) We can still consider the Q~-cohomology in the space of operators that are annihilated by J ′. The vertex algebra V ~ is the Q~-cohomology of twisted-translated ′ ~ ~ ~ local operators, placed at w = 0 and annihilated by J . Since Pz¯ is Q -exact, classes in V vary holomorphicallye on C, just as in the undeformed case. Unlike the undeformed case, ~ ~ e however, the product [[O1(z1)]][[O2 (z2)]] of two local Q-cohomology classes can be singular at z1 = z2 because operators representing these classes are located at the same point w = 0 on R2. The structure of V ~ is most naturally described in the Kapustin twist [31], in which the rotation generators (J, J) are replaced by (J ′, J ′), where J ′ = J + R/2 as in the case of three-dimensional N = 2 supersymmetry. In the Kapustin twist, Q transforms under rotations as a scalar and eP ~, Q as one-forms. Thee R-grading is well defined in V ~ if we assign R(~) = 2 , (4.30) for then Q~ has a definite charge R(Q~) = 1, while R(P ~) = 0 and the twisted translation preserves R. The operator product expansion (OPE) takes the form

c ~ ~ ~ ~ Cab ( )[[Oc (z2)]] [[Oa(z1)]][[Ob (z2)]] = , (4.31) (z − z )ha+hb−hc c 1 2 X ′ c where ha = J (Oa) and Cab (~) are analytic functions of ~. Having a singular OPE struc- ture, V ~ is a vertex algebra, rather than a Poisson vertex algebra. For a unitary N = 2 superconformal field theory, it was shown in [25] that V ~ is isomorphic to the vertex operator algebra introduced in [7]. The latter is the cohomology 1 ~ 2−˙ of twisted-translated local operators with respect to the supercharge Q− + S , which is “half” of Q~, so to speak. This supercharge has r = 1/2, so V ~ has an additional grading by r in the unitary case.

– 21 – ~ ~ ~ We can encode the singular part of the OPE in the λ-bracket [ λ ]: V ⊗ V → V [λ] on V ~, defined by

~ ~ dz1 λ(z1−z2) ~ ~ [[[O1]] λ [[O2]]](z2)= e [[O1(z1)]][[O2 (z2)]] S1 2πi Z z2 ∞ n (4.32) λ ~ = δ C c(~)[[O (z )]] . n! h1+h2−hc,n+1 12 c 2 n c X=0 X The λ-bracket may be considered for any vertex algebra, but that of V ~ has a special property. In the limit ~ → 0, nontrivial Q~-cohomology classes reduce to nontrivial Q- cohomology classes; the ~-correction to Q can destroy but not create cohomology classes, as we will argue shortly. Since the OPE between Q-cohomology classes is regular, the OPE c coefficients Cab (~) for ha +hb −hc > 0 contain only terms of positive order when expanded in powers of ~. In particular, the λ-bracket is at least of order ~.8 It is known that in such a situation, the family of vertex algebras V ~ reduces in the limit ~ → 0 to a Poisson vertex algebra V′; see [19] for a proof. The λ-bracket of V′ is given by 1 ~ ~ {[[O1]] λ [[O2]]} = lim [[[O1]] λ [[O2 ]]] . (4.33) ~→0 ~ The appearance of Poisson vertex algebras in this context was previously noted in [32, 33]. Now we show that V′ is a subalgebra of the Poisson vertex algebra V associated with the topological–holomorphic sector of the theory. First of all, we note that it makes sense to compare V ~ and V as vector spaces. Since J ′ is Q-exact, for the computation of the Q-cohomology of local operators we can restrict Q to the kernel of J ′. We can also compute it anywhere on R2. Thus we can compute V, ase a vector space, as the Q-cohomology in the space of of local operators that have J ′ = 0 and are located at we = 0. This is the space in which the Q~-cohomology of local operators is defined. e An injection V ~ →V is constructed as follows. Let [[O~]] be a nontrivial Q~-cohomology class. We may assume that [[O~]] neither vanishes nor diverges in the limit ~ → 0; we can ~ ~ multiply [[O ]] by an appropriate power of ~ if needed. Thus we have O = O + ~O1, ~ with O containing no ~ and O1 nonnegative powers of ~. If we write Q = Q + ~Q1, then the relation (Q~)2 = 0 (which holds in the space of operators under consideration) implies 2 ~ ~ ~ [Q,Q1]= Q1 = 0, while Q · O = 0 implies Q · O = Q1 · O + Q · O1 + Q1 · O1 = 0. Hence, O represents a Q-cohomology class. This class is nontrivial because if O = Q · O′ for some ′ ~ ′ ~ ~ ′ ′ local operator O , then Q · (O1 − Q1 · O ) = 0 and O = Q · O + ~(O1 − Q1 · O ), in contradiction with the assumption.

8In [25], the construction of vertex algebras was extended to = 2 supersymmetric gauge theories which N are not necessarily conformal. For a nonconformal theory, the associated vertex algebra is anomalous, but the anomaly can be canceled if an = (0, 2) supersymmetric surface defect carrying an appropriate vertex N algebra is inserted at w = 0. Although the combined vertex algebra is well-defined, it may not have a classical limit since the vertex algebra on the surface defect may not have one. What goes wrong in the above argument is the assertion that the OPE between Q-cohomology classes is regular.

– 22 – To show that the λ-bracket of V ~ reduces to that of V in the limit ~ → 0, we introduce an equivariant analog of topological–holomorphic descent. The key equation (2.12) is deformed to

~ ~ ±∗ ′ ~ ~ ±∗ Q · O (w, w,z,¯ z¯) = ±(d + ιJe) O (w, w,z,¯ z¯) ~ ~ ~ ∓∗
+ exp(iwPw + iwP ¯
w¯) · (Q · O(0, 0, z, z¯)) , (4.34)

For a local operator O that is Q~-closed at w = 0, we have

~ ~ ~ ~ Q · (ω ∧ (O )∗)= −d (ω ∧ (O )∗) (4.35) for any holomorphic one-form ω on C, where

~ ~ d =d+ ιJe . (4.36)

Compared to the relation (2.14), the exterior derivative d is deformed to its equivariant version d~, which squares to the Lie derivative by ~J. Accordingly, we can integrate ω ∧ (O~)∗ over an equivariant cycle Γ~ to obtain a Q~- closed operator, and the Q~-cohomology class of thee resulting operator depends only on [[O~]] ∈ V ~ and the equivariant homology class [[Γ~]]. The equivariant homology is the homology with respect to the boundary operator

~ ∂ = ∂ + ~J , (4.37) where J is the dual of ιJe. Acting on a k-chain located at some fixed value of ϕ = arg w, the operator J sweeps it around in the ϕ-direction to produce a (k + 1)-chain that is invariant under the action of J = −i∂ϕ. Let us consider the Q~-cohomology class e

λ(z1−z2) ~ ∗ ~ e dz1 ∧ (O1) (x1) O2(x2) . (4.38) S3 Z x2   2 ~ 2 1 ~ 3 1 There is a 2-disk Dx2 such that ∂ Dx2 = Sz2 + Sx2 /2πi, with Sz2 being a circle lying in 9 C, centered at z2 and located at w = 0. Hence, we have 2πi [[S3 ]] = − [[S1 ]] (4.40) x2 ~ z2

9In terms of the spherical coordinates (ψ,θ,ϕ) [0,π] [0,π] [0, 2π), defined by ∈ × × 1 1 x1 x2 = r sin ψ sin θ cos ϕ , 2 − 2 x1 x2 = r sin ψ sin θ sin ϕ , − 3 3 (4.39) x1 x2 = r cos ψ , 4 − 4 x1 x2 = r sin ψ cos θ , − 2 1 the 2-disk Dx2 is located at ϕ = 0 and has the boundary Sz2 at θ = 0, π. One can easily show that for an 3 1 equivariantly closed form, the integral over Sx2 reduces to an integral over Sz2 .

– 23 – in the equivariant homology, and the above Q~-cohomology class is equal, up to an overall ~ ~ ~ ~ numerical factor, to [[[O1 ]] λ [[O2]]](z2)/ . In the limit → 0, the expression (4.38) reduces to the λ-bracket {[[O1]] λ [[O2]]}(z2), leading to the desired relation between the two λ-brackets. If the theory is unitary, the limit ~ → 0 of V ~ equals V and not a proper subalgebra thereof. In this case, the elements of V ~ are in one-to-one correspondence with the harmonic states of Q~, that is, those states satisfying the condition

~ ~ [Q , (Q )†]=(1+ ~2)(D − J − R) = 0 , (4.41) where D is the dilatation operator. This condition is independent of ~, hence equivalent to the harmonic condition [Q,Q†] = 0 characterizing the elements of V. The vacuum character of V ~ computes the Schur limit of the superconformal index of the theory [7]. Since the character is independent of ~, the character of V also equals the Schur index. The Poisson vertex algebra can be defined for nonconformal theories, and its character may be taken as a definition of the Schur index for those theories.

4.3 Free hypermultiplets Let us determine the Poisson vertex algebra for the theory of a free hypermultiplet. To this purpose it is convenient to adopt a two-dimensional point of view. 1 2 The supercharges Q−, Q2−˙ and Q+, Q1+˙ which comprise Q and (part of) Q form a subalgebra isomorphic to the N = (2, 2) supersymmetry algebra on R2. The latter may be obtained by dimensional reduction from the N = 1 supersymmetry algebra in four dimensions or the N = 2 supersymmetry algebra in three dimensions. It is generated by four supercharges Q±, Q±, satisfying

[Q+, Q+]= Pw , [Q−, Q−]= −Pw¯ .

[Q+, Q−]= Z , [Q−, Q+]= Z , (4.42)

[Q+,Q−] = 0 , [Q+, Q−] = 0 , where Z is a complex central charge. We have the identification

1 2 Q− = Q− , Q2−˙ = Q+ , Q+ = Q+ , Q1+˙ = −Q− , (4.43) and Z coincides with the central charge in the four-dimensional N = 2 supersymmetry algebra. (In general, the N = (2, 2) supersymmetry algebra has an additional central charge, which is zero in the present case.) We may therefore think of any N = 2 supersymmetric field theory on R2 × C as an N = (2, 2) supersymmetric field theory on R2. From this two-dimensional point of view, a hypermultiplet consists of a pair of N = (2, 2) chiral multiplets, which have “continuous indices” (z, z¯), namely their coordinates on C. A chiral multiplet Φ of N = (2, 2) supersymmetry is the dimensional reduction of a chiral multiplet of N = 2 supersymmetry in three dimensions. Under the action of

– 24 – ǫ−Q+ − ǫ+Q− − ǫ¯−Q+ +ǫ ¯+Q−, the fields of Φ transform as

δφ = ǫ−ψ+ − ǫ+ψ− ,

δψ+ = i¯ǫ−∂wφ +ǫ ¯+mφ + ǫ+F , (4.44) δψ− = −ǫ¯−mφ¯ + i¯ǫ+∂w¯φ + ǫ−F ,

δF = −i¯ǫ+(∂w¯ψ+ + imψ−) − i¯ǫ−(imψ ¯ + − ∂wψ−) , with m a complex mass parameter, the twisted mass of Φ. The central charge is given by

Z(Φ) = m . (4.45)

Let us rename the fields of Φ and its conjugate Φ as

χ = iψ+dw − iψ−dw ¯ , G = F dw ∧ dw ¯ , 1 (4.46) η¯ = ψ¯ + ψ¯ , µ¯ = (ψ¯ − ψ¯ ) . + − 2 + −

Then, the action of Q = Q− + Q+ on the fields can be written as

Q · φ = 0 , Q · φ¯ =η ¯ ,

Q · χ =dmφ, Q · η¯ = 0 , (4.47) Q · G =dmχ, Q · µ¯ = F , Q · F = 0 , where

dm =dw(∂w + im)+d¯w(∂w¯ − im ¯ ) (4.48) is the exterior derivative twisted by the twisted mass m. As in the case of N = 2 super- symmetry in three dimensions, in terms of the twisted-translated fields the twisted mass disappears from the formula. a a a a We are interested in a theory consisting of multiple chiral multiplets Φ = (φ , ψ±, F ), coupled through a superpotential W , which is a holomorphic function of φa. Integrating out the auxiliary fields sets

¯ ∂W a¯ ∂W F a = −gab , F = −gab¯ , (4.49) ∂φ¯¯b ∂φb

a¯b ab¯ a¯ with g = g = δab. For a generic superpotential, the fields of the antichiral multiplets Φ do not contribute to the Q-cohomology. Then, the Q-cohomology is spanned (at w = 0) by holomorphic functions of φa, but there is the relation

dW = 0 (4.50) coming from Q · µ¯a¯. Modulo d¯z, Q acts on Φa as 1 Q · φa = χa , Q · χa = Ga , (4.51) 2

– 25 – from which it follows (φa)∗ = φa + χa + Ga . (4.52) A hypermultiplet on R2 × C consists of a pair of chiral multiplets with opposite masses m and −m, whose scalar fields qa(z, z¯), a = 1, 2, have (R, J ′) = (1, 1/2) and are coupled by the superpotential a b W ∝ dz ∧ d¯zǫabq ∂z¯q . (4.53) ZC The Q-cohomology of local operators is the algebra of holomorphic functions of qa, and by the relation (4.50) its classes vary holomorphically on C, as they should. ′ a b The λ-bracket has (R, J ) = (−2, −1), hence {[[q ]] λ [[q ]]} is a Q-cohomology class of (R, J ′) = (0, 0). Furthermore, it should be invariant under the SU(2) symmetry rotating (q1,q2) as a doublet. The only possibility is a multiple of ǫab. Let us show this explicitly. As explained near the end of section 4.1, we can compute the λ-bracket by taking the R2 1 integration cycle to be × Sz2 :

a b λ(z1−z2) Z a ∗ b {[[q ]] λ [[q ]]}(z2)= e dz1 ∧ ( q ) (x1) q (z2) . (4.54) R2×S1 Z z2   We have

Z a ∗ b imw1−im ¯ w¯1 ac¯ d¯ b ( q ) (x1) q (0) ∝ e g ǫc¯d¯∂zq¯ (x1)dw1 ∧ dw ¯1 q (x2) , (4.55) R2 R2 Z  Z  which, using the propagator, we can rewrite as µ µ −ipµ(x1 −x2 ) imw1−im ¯ w¯1 ab 4 e e ǫ ∂z1 d p µ 2 dw1 ∧ dw ¯1 . (4.56) R2 R4 p p + |m| Z Z µ  Integrating over w1,w ¯1 yields delta functions which set pw = m and pw¯ = −m¯ , leaving

−ipz(z1−z2)−ipz¯(¯z1−z¯2) ab ab 2 e ǫ ǫ ∂z1 d p ∝ . (4.57) R2 p p z − z Z z z¯  1 2 The last proportionality can be seen from the fact that acted on with ∂z¯1 , both sides become 1 a delta function supported at z1 = z2. Performing the integral over Sz2 , we find a b ab {[[q ]] λ [[q ]]}∝ ǫ , (4.58) as expected. The result is also independent of m due to the locality of V, as explained in section 3.2. For m = 0, the hypermultiplet is conformal. The associated vertex algebra is known to be the algebra of symplectic bosons [7], characterized by the OPE ab a ~ b ~ ǫ [[(q ) (z1)]][[(q ) (z2)]] ∼ ~ . (4.59) z1 − z2 In the classical limit ~ → 0, this vertex algebra indeed reduces to the Poisson vertex algebra just found. 2 The Poisson algebra V/(∂zV)V is isomorphic to that of holomorphic functions on C a b with respect to the holomorphic symplectic form ǫab dq ∧ dq . This is the Poisson algebra associated with the Rozansky–Witten twist [34] of a three-dimensional N = 4 hypermulti- plet [6].

– 26 – 4.4 Gauge theories Finally, let us determine the Poisson vertex algebra for an N = 2 supersymmetric gauge theory, constructed from a vector multiplet for a gauge group G and a hypermultiplet in a representation ρ of G. If ρ is chosen appropriately and the hypermultiplet masses are zero, the theory is conformal. In this case the associated vertex algebra can be defined, and it is known to be the algebra of gauged symplectic bosons [7, 25, 26]. This algebra is constructed as follows. Let γ and β be bosonic fields of conformal weight J ′ = 1/2, valued in ρ and its dual ρ∨, and b and c be fermionic fields with J ′ = 1 and 0, valued in the adjoint representation of G. The ghost field c is constrained to have no zero mode. The dynamics of these fields are governed by the action 1 ~ dz ∧ d¯z(b∂z¯c + β∂z¯γ) , (4.60) ZC which lead to the OPEs

~ idρ ~C2(g) γ(z1)β(z2) ∼ , b(z1)c(z2) ∼ . (4.61) z1 − z2 z1 − z2

Here idρ is the identity operator on the representation space of ρ, and C2(g) is the quadratic ~ Casimir element of the Lie algebra g of G. Let Vbc–βγ be the vertex algebra generated by β, γ, b, c. ~ The vertex algebra Vbc–βγ has a fermionic symmetry, called the BRST symmetry, whose conserved current is given by

JBRST = Tr(bcc) − βcγ . (4.62)

~ The corresponding charge QBRST acts on an element O ∈ Vbc–βγ by

1 QBRST · O(z)= JBRST O(z) (4.63) 2πi~ 1 ZSz  2 and satisfies QBRST = 0. The vertex algebra of the gauge theory is the QBRST-cohomology ~ of Vbc–βγ . The vertex algebra for a free massless hypermultiplet corresponds to the case when G is trivial and ([[q1]], [[q2]]) = (γ, β). The Poisson vertex algebra associated with the gauge theory is the classical limit of ~ the BRST cohomology, and can be described as follows. The vertex algebra Vbc–βγ reduces to a Poisson vertex algebra Vbc–βγ whose λ-bracket is given by

{γ λ β}∝ idρ , {b λ c}∝ C2(g) . (4.64) ~ ~ The action (4.63) of QBRST on Vbc–βγ is given by 1/ times [JBRST λ ]|λ=0. In the limit ~ → 0, this becomes the action of the zero mode JBRST on Vbc–βγ. Therefore, the Poisson vertex algebra for the N = 2 superconformal gauge theory is the classical BRST R cohomology of Vbc–βγ , whose differential is given by { JBRST, }. The classical BRST cohomology of V actually makes sense for any choice of ρ, bc–βγ R not necessarily one for which the theory is conformal. This is in contrast to its quantum

– 27 – counterpart: QBRST squares to zero if and only if the one-loop beta function vanishes. The anomaly in the BRST symmetry arises from double contractions in the OPE between two 2 JBRST, which is of order ~ and discarded in the classical limit. Based on this observation, we propose that the Poisson vertex algebra for an N = 2 supersymmetric gauge theory, whether conformal or not, is given by the same classical BRST cohomology. The character of this Poisson vertex algebra can be calculated by a matrix integral for- mula. In [17], this formula was employed as a definition of the Schur index for nonconformal theories with Lagrangian descriptions, and shown to coincide with a certain wall-crossing invariant in a number of examples.

Acknowledgments

We would like to thank Kevin Costello for invaluable advice and illuminating discussions, and Dylan Butson, Tudor Dimofte and Davide Gaiotto for helpful comments. The research of JO is supported in part by Kwanjeong Educational Foundation, by the Visiting Grad- uate Fellowship Program at the Perimeter Institute for Theoretical Physics, and by the Berkeley Center of Theoretical Physics. The research of JY is supported by the Perimeter Institute for Theoretical Physics. Research at Perimeter Institute is supported in part by the Government of Canada through the Department of Innovation, Science and Economic Development Canada and by the Province of Ontario through the Ministry of Colleges and Universities.

References

[1] A. Beilinson and V. Drinfeld, Chiral algebras, vol. 51 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 2004. [2] B. Enriquez and E. Frenkel, Geometric interpretation of the Poisson structure in affine Toda field theories, Duke Math. J. 92 (1998) 459–495 [q-alg/9606023]. [3] E. Frenkel and D. Ben-Zvi, Vertex algebras and algebraic curves, vol. 88 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, second ed., 2004. 10.1090/surv/088. [4] E. Witten, Topological quantum field theory, Comm. Math. Phys. 117 (1988) 353–386. [5] E. Witten, Topological sigma models, Comm. Math. Phys. 118 (1988) 411–449. [6] C. Beem, D. Ben-Zvi, M. Bullimore, T. Dimofte and A. Neitzke, Secondary products in supersymmetric field theory, Ann. Henri Poincar´e 21 (2020) 1235–1310 [1809.00009]. [7] C. Beem, M. Lemos, P. Liendo, W. Peelaers, L. Rastelli and B. C. van Rees, Infinite chiral symmetry in four dimensions, Comm. Math. Phys. 336 (2015) 1359–1433 [1312.5344]. [8] V. Kac, Vertex algebras for beginners, vol. 10 of University Lecture Series. American Mathematical Society, Providence, RI, second ed., 1998. 10.1090/ulect/010. [9] P. Safronov, Braces and Poisson additivity, Compos. Math. 154 (2018) 1698–1745 [1611.09668].

– 28 – [10] K. Costello, T. Dimofte and D. Gaiotto, Boundary chiral algebras and holomorphic twists, 2005.00083. [11] D. Gaiotto, N =2 dualities, JHEP 08 (2012) 034 [0904.2715]. [12] D. Gaiotto, G. W. Moore and A. Neitzke, Wall-crossing, Hitchin systems, and the WKB approximation, Adv. Math. 234 (2013) 239–403 [0907.3987]. [13] Y. Terashima and M. Yamazaki, SL(2, R) Chern–Simons, Liouville, and gauge theory on duality walls, JHEP 08 (2011) 135 [1103.5748]. [14] Y. Terashima and M. Yamazaki, Semiclassical analysis of the 3d/3d relation, Phys. Rev. D 88 (2013) 026011 [1106.3066]. [15] T. Dimofte, D. Gaiotto and S. Gukov, Gauge theories labelled by three-manifolds, Comm. Math. Phys. 325 (2014) 367 [1108.4389]. [16] T. Dimofte, D. Gaiotto and S. Gukov, 3-manifolds and 3d indices, Adv. Theor. Math. Phys. 17 (2013) 975 [1112.5179]. [17] C. C´ordova and S.-H. Shao, Schur indices, BPS particles, and Argyres–Douglas theories, JHEP 01 (2016) 040 [1506.00265]. [18] A. Barakat, A. De Sole and V. G. Kac, Poisson vertex algebras in the theory of Hamiltonian equations, Jpn. J. Math. 4 (2009) 141–252. [19] V. Kac, Introduction to vertex algebras, Poisson vertex algebras, and integrable Hamiltonian PDE, in Perspectives in Lie theory, vol. 19 of Springer INdAM Ser., pp. 3–72. Springer, Cham, 2017. 1512.00821. [20] J. Wess and J. Bagger, Supersymmetry and supergravity. Princeton Series in Physics. Princeton University Press, Princeton, NJ, second ed., 1992. [21] C. C´ordova, T. T. Dumitrescu and K. Intriligator, Multiplets of superconformal symmetry in diverse dimensions, JHEP 03 (2019) 163 [1612.00809]. [22] C. Beem, “Comments on vertex algebras for N = 2 SCFTs.” Talk at String Math 2017, July 24–28, 2017. [23] C. Beem, “4d N = 2 SCFTs and VOAs.” Talk at Pollica Summer Workshop “Mathematical and Geometric Tools for Conformal Field Theories,” June 3–21, 2019. [24] C. Beem, “Building VOAs out of Higgs branches.” Talk at String Math 2019, July 1–5, 2019. [25] J. Oh and J. Yagi, Chiral algebras from Ω-deformation, JHEP 08 (2019) 143 [1903.11123]. [26] S. Jeong, SCFT/VOA correspondence via Ω-deformation, JHEP 10 (2019) 171 [1904.00927]. [27] C. Beem, W. Peelaers and L. Rastelli, Deformation quantization and superconformal symmetry in three dimensions, Comm. Math. Phys. 354 (2017) 345–392 [1601.05378]. [28] N. A. Nekrasov, Seiberg–Witten prepotential from instanton counting, Adv. Theor. Math. Phys. 7 (2003) 831–864,[hep-th/0206161]. [29] N. A. Nekrasov and A. Okounkov, Seiberg–Witten theory and random partitions, in The unity of mathematics, vol. 244 of Progr. Math., p. 525. Birkh¨auser Boston, Boston, MA, 2006. hep-th/0306238. DOI. [30] J. Yagi, Ω-deformation and quantization, JHEP 08 (2014) 112 [1405.6714].

– 29 – [31] A. Kapustin, Holomorphic reduction of N =2 gauge theories, Wilson–’t Hooft operators, and S-duality, hep-th/0612119. [32] C. Beem and L. Rastelli, Vertex operator algebras, Higgs branches, and modular differential equations, JHEP 08 (2018) 114 [1707.07679]. [33] C. Beem, C. Meneghelli and L. Rastelli, Free field realizations from the Higgs branch, JHEP 09 (2019) 058 [1903.07624]. [34] L. Rozansky and E. Witten, Hyper-K¨ahler geometry and invariants of three-manifolds, Selecta Math. (N.S.) 3 (1997) 401 [hep-th/9612216].

– 30 –